FFmpeg
vf_dnn_processing.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019 Guo Yejun
3  *
4  * This file is part of FFmpeg.
5  *
6  * FFmpeg is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * FFmpeg is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with FFmpeg; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20 
21 /**
22  * @file
23  * implementing a generic image processing filter using deep learning networks.
24  */
25 
26 #include "libavformat/avio.h"
27 #include "libavutil/opt.h"
28 #include "libavutil/pixdesc.h"
29 #include "libavutil/avassert.h"
30 #include "libavutil/imgutils.h"
31 #include "avfilter.h"
32 #include "dnn_interface.h"
33 #include "formats.h"
34 #include "internal.h"
35 #include "libswscale/swscale.h"
36 
37 typedef struct DnnProcessingContext {
38  const AVClass *class;
39 
44 
47 
48  // input & output of the model at execution time
51 
57 
58 #define OFFSET(x) offsetof(DnnProcessingContext, x)
59 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_VIDEO_PARAM
60 static const AVOption dnn_processing_options[] = {
61  { "dnn_backend", "DNN backend", OFFSET(backend_type), AV_OPT_TYPE_INT, { .i64 = 0 }, INT_MIN, INT_MAX, FLAGS, "backend" },
62  { "native", "native backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "backend" },
63 #if (CONFIG_LIBTENSORFLOW == 1)
64  { "tensorflow", "tensorflow backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "backend" },
65 #endif
66 #if (CONFIG_LIBOPENVINO == 1)
67  { "openvino", "openvino backend flag", 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, 0, 0, FLAGS, "backend" },
68 #endif
69  { "model", "path to model file", OFFSET(model_filename), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
70  { "input", "input name of the model", OFFSET(model_inputname), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
71  { "output", "output name of the model", OFFSET(model_outputname), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
72  { NULL }
73 };
74 
75 AVFILTER_DEFINE_CLASS(dnn_processing);
76 
78 {
79  DnnProcessingContext *ctx = context->priv;
80 
81  if (!ctx->model_filename) {
82  av_log(ctx, AV_LOG_ERROR, "model file for network is not specified\n");
83  return AVERROR(EINVAL);
84  }
85  if (!ctx->model_inputname) {
86  av_log(ctx, AV_LOG_ERROR, "input name of the model network is not specified\n");
87  return AVERROR(EINVAL);
88  }
89  if (!ctx->model_outputname) {
90  av_log(ctx, AV_LOG_ERROR, "output name of the model network is not specified\n");
91  return AVERROR(EINVAL);
92  }
93 
95  if (!ctx->dnn_module) {
96  av_log(ctx, AV_LOG_ERROR, "could not create DNN module for requested backend\n");
97  return AVERROR(ENOMEM);
98  }
99  if (!ctx->dnn_module->load_model) {
100  av_log(ctx, AV_LOG_ERROR, "load_model for network is not specified\n");
101  return AVERROR(EINVAL);
102  }
103 
104  ctx->model = (ctx->dnn_module->load_model)(ctx->model_filename);
105  if (!ctx->model) {
106  av_log(ctx, AV_LOG_ERROR, "could not load DNN model\n");
107  return AVERROR(EINVAL);
108  }
109 
110  return 0;
111 }
112 
114 {
115  static const enum AVPixelFormat pix_fmts[] = {
121  };
122  AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
123  return ff_set_common_formats(context, fmts_list);
124 }
125 
126 #define LOG_FORMAT_CHANNEL_MISMATCH() \
127  av_log(ctx, AV_LOG_ERROR, \
128  "the frame's format %s does not match " \
129  "the model input channel %d\n", \
130  av_get_pix_fmt_name(fmt), \
131  model_input->channels);
132 
133 static int check_modelinput_inlink(const DNNData *model_input, const AVFilterLink *inlink)
134 {
135  AVFilterContext *ctx = inlink->dst;
136  enum AVPixelFormat fmt = inlink->format;
137 
138  // the design is to add explicit scale filter before this filter
139  if (model_input->height != -1 && model_input->height != inlink->h) {
140  av_log(ctx, AV_LOG_ERROR, "the model requires frame height %d but got %d\n",
141  model_input->height, inlink->h);
142  return AVERROR(EIO);
143  }
144  if (model_input->width != -1 && model_input->width != inlink->w) {
145  av_log(ctx, AV_LOG_ERROR, "the model requires frame width %d but got %d\n",
146  model_input->width, inlink->w);
147  return AVERROR(EIO);
148  }
149 
150  switch (fmt) {
151  case AV_PIX_FMT_RGB24:
152  case AV_PIX_FMT_BGR24:
153  if (model_input->channels != 3) {
155  return AVERROR(EIO);
156  }
157  if (model_input->dt != DNN_FLOAT && model_input->dt != DNN_UINT8) {
158  av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type as float32 and uint8.\n");
159  return AVERROR(EIO);
160  }
161  return 0;
162  case AV_PIX_FMT_GRAY8:
163  if (model_input->channels != 1) {
165  return AVERROR(EIO);
166  }
167  if (model_input->dt != DNN_UINT8) {
168  av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type uint8.\n");
169  return AVERROR(EIO);
170  }
171  return 0;
172  case AV_PIX_FMT_GRAYF32:
173  case AV_PIX_FMT_YUV420P:
174  case AV_PIX_FMT_YUV422P:
175  case AV_PIX_FMT_YUV444P:
176  case AV_PIX_FMT_YUV410P:
177  case AV_PIX_FMT_YUV411P:
178  if (model_input->channels != 1) {
180  return AVERROR(EIO);
181  }
182  if (model_input->dt != DNN_FLOAT) {
183  av_log(ctx, AV_LOG_ERROR, "only support dnn models with input data type float32.\n");
184  return AVERROR(EIO);
185  }
186  return 0;
187  default:
188  av_log(ctx, AV_LOG_ERROR, "%s not supported.\n", av_get_pix_fmt_name(fmt));
189  return AVERROR(EIO);
190  }
191 
192  return 0;
193 }
194 
196 {
197  AVFilterContext *context = inlink->dst;
198  DnnProcessingContext *ctx = context->priv;
200  DNNData model_input;
201  int check;
202 
203  result = ctx->model->get_input(ctx->model->model, &model_input, ctx->model_inputname);
204  if (result != DNN_SUCCESS) {
205  av_log(ctx, AV_LOG_ERROR, "could not get input from the model\n");
206  return AVERROR(EIO);
207  }
208 
209  check = check_modelinput_inlink(&model_input, inlink);
210  if (check != 0) {
211  return check;
212  }
213 
214  ctx->input.width = inlink->w;
215  ctx->input.height = inlink->h;
216  ctx->input.channels = model_input.channels;
217  ctx->input.dt = model_input.dt;
218 
219  result = (ctx->model->set_input_output)(ctx->model->model,
220  &ctx->input, ctx->model_inputname,
221  (const char **)&ctx->model_outputname, 1);
222  if (result != DNN_SUCCESS) {
223  av_log(ctx, AV_LOG_ERROR, "could not set input and output for the model\n");
224  return AVERROR(EIO);
225  }
226 
227  return 0;
228 }
229 
230 static int prepare_sws_context(AVFilterLink *outlink)
231 {
232  AVFilterContext *context = outlink->src;
233  DnnProcessingContext *ctx = context->priv;
234  AVFilterLink *inlink = context->inputs[0];
235  enum AVPixelFormat fmt = inlink->format;
236  DNNDataType input_dt = ctx->input.dt;
237  DNNDataType output_dt = ctx->output.dt;
238 
239  switch (fmt) {
240  case AV_PIX_FMT_RGB24:
241  case AV_PIX_FMT_BGR24:
242  if (input_dt == DNN_FLOAT) {
243  ctx->sws_gray8_to_grayf32 = sws_getContext(inlink->w * 3,
244  inlink->h,
246  inlink->w * 3,
247  inlink->h,
249  0, NULL, NULL, NULL);
250  }
251  if (output_dt == DNN_FLOAT) {
252  ctx->sws_grayf32_to_gray8 = sws_getContext(outlink->w * 3,
253  outlink->h,
255  outlink->w * 3,
256  outlink->h,
258  0, NULL, NULL, NULL);
259  }
260  return 0;
261  case AV_PIX_FMT_YUV420P:
262  case AV_PIX_FMT_YUV422P:
263  case AV_PIX_FMT_YUV444P:
264  case AV_PIX_FMT_YUV410P:
265  case AV_PIX_FMT_YUV411P:
266  av_assert0(input_dt == DNN_FLOAT);
267  av_assert0(output_dt == DNN_FLOAT);
268  ctx->sws_gray8_to_grayf32 = sws_getContext(inlink->w,
269  inlink->h,
271  inlink->w,
272  inlink->h,
274  0, NULL, NULL, NULL);
275  ctx->sws_grayf32_to_gray8 = sws_getContext(outlink->w,
276  outlink->h,
278  outlink->w,
279  outlink->h,
281  0, NULL, NULL, NULL);
282 
283  if (inlink->w != outlink->w || inlink->h != outlink->h) {
285  int sws_src_h = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
286  int sws_src_w = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
287  int sws_dst_h = AV_CEIL_RSHIFT(outlink->h, desc->log2_chroma_h);
288  int sws_dst_w = AV_CEIL_RSHIFT(outlink->w, desc->log2_chroma_w);
289  ctx->sws_uv_scale = sws_getContext(sws_src_w, sws_src_h, AV_PIX_FMT_GRAY8,
290  sws_dst_w, sws_dst_h, AV_PIX_FMT_GRAY8,
292  ctx->sws_uv_height = sws_src_h;
293  }
294  return 0;
295  default:
296  //do nothing
297  break;
298  }
299 
300  return 0;
301 }
302 
303 static int config_output(AVFilterLink *outlink)
304 {
305  AVFilterContext *context = outlink->src;
306  DnnProcessingContext *ctx = context->priv;
308 
309  // have a try run in case that the dnn model resize the frame
310  result = (ctx->dnn_module->execute_model)(ctx->model, &ctx->output, 1);
311  if (result != DNN_SUCCESS){
312  av_log(ctx, AV_LOG_ERROR, "failed to execute model\n");
313  return AVERROR(EIO);
314  }
315 
316  outlink->w = ctx->output.width;
317  outlink->h = ctx->output.height;
318 
319  prepare_sws_context(outlink);
320 
321  return 0;
322 }
323 
325 {
326  int bytewidth = av_image_get_linesize(frame->format, frame->width, 0);
327  DNNData *dnn_input = &ctx->input;
328 
329  switch (frame->format) {
330  case AV_PIX_FMT_RGB24:
331  case AV_PIX_FMT_BGR24:
332  if (dnn_input->dt == DNN_FLOAT) {
333  sws_scale(ctx->sws_gray8_to_grayf32, (const uint8_t **)frame->data, frame->linesize,
334  0, frame->height, (uint8_t * const*)(&dnn_input->data),
335  (const int [4]){frame->width * 3 * sizeof(float), 0, 0, 0});
336  } else {
337  av_assert0(dnn_input->dt == DNN_UINT8);
338  av_image_copy_plane(dnn_input->data, bytewidth,
339  frame->data[0], frame->linesize[0],
340  bytewidth, frame->height);
341  }
342  return 0;
343  case AV_PIX_FMT_GRAY8:
344  case AV_PIX_FMT_GRAYF32:
345  av_image_copy_plane(dnn_input->data, bytewidth,
346  frame->data[0], frame->linesize[0],
347  bytewidth, frame->height);
348  return 0;
349  case AV_PIX_FMT_YUV420P:
350  case AV_PIX_FMT_YUV422P:
351  case AV_PIX_FMT_YUV444P:
352  case AV_PIX_FMT_YUV410P:
353  case AV_PIX_FMT_YUV411P:
354  sws_scale(ctx->sws_gray8_to_grayf32, (const uint8_t **)frame->data, frame->linesize,
355  0, frame->height, (uint8_t * const*)(&dnn_input->data),
356  (const int [4]){frame->width * sizeof(float), 0, 0, 0});
357  return 0;
358  default:
359  return AVERROR(EIO);
360  }
361 
362  return 0;
363 }
364 
366 {
367  int bytewidth = av_image_get_linesize(frame->format, frame->width, 0);
368  DNNData *dnn_output = &ctx->output;
369 
370  switch (frame->format) {
371  case AV_PIX_FMT_RGB24:
372  case AV_PIX_FMT_BGR24:
373  if (dnn_output->dt == DNN_FLOAT) {
374  sws_scale(ctx->sws_grayf32_to_gray8, (const uint8_t *[4]){(const uint8_t *)dnn_output->data, 0, 0, 0},
375  (const int[4]){frame->width * 3 * sizeof(float), 0, 0, 0},
376  0, frame->height, (uint8_t * const*)frame->data, frame->linesize);
377 
378  } else {
379  av_assert0(dnn_output->dt == DNN_UINT8);
380  av_image_copy_plane(frame->data[0], frame->linesize[0],
381  dnn_output->data, bytewidth,
382  bytewidth, frame->height);
383  }
384  return 0;
385  case AV_PIX_FMT_GRAY8:
386  // it is possible that data type of dnn output is float32,
387  // need to add support for such case when needed.
388  av_assert0(dnn_output->dt == DNN_UINT8);
389  av_image_copy_plane(frame->data[0], frame->linesize[0],
390  dnn_output->data, bytewidth,
391  bytewidth, frame->height);
392  return 0;
393  case AV_PIX_FMT_GRAYF32:
394  av_assert0(dnn_output->dt == DNN_FLOAT);
395  av_image_copy_plane(frame->data[0], frame->linesize[0],
396  dnn_output->data, bytewidth,
397  bytewidth, frame->height);
398  return 0;
399  case AV_PIX_FMT_YUV420P:
400  case AV_PIX_FMT_YUV422P:
401  case AV_PIX_FMT_YUV444P:
402  case AV_PIX_FMT_YUV410P:
403  case AV_PIX_FMT_YUV411P:
404  sws_scale(ctx->sws_grayf32_to_gray8, (const uint8_t *[4]){(const uint8_t *)dnn_output->data, 0, 0, 0},
405  (const int[4]){frame->width * sizeof(float), 0, 0, 0},
406  0, frame->height, (uint8_t * const*)frame->data, frame->linesize);
407  return 0;
408  default:
409  return AVERROR(EIO);
410  }
411 
412  return 0;
413 }
414 
416 {
417  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt);
418  av_assert0(desc);
419  return !(desc->flags & AV_PIX_FMT_FLAG_RGB) && desc->nb_components == 3;
420 }
421 
423 {
424  const AVPixFmtDescriptor *desc;
425  int uv_height;
426 
427  if (!ctx->sws_uv_scale) {
428  av_assert0(in->height == out->height && in->width == out->width);
429  desc = av_pix_fmt_desc_get(in->format);
430  uv_height = AV_CEIL_RSHIFT(in->height, desc->log2_chroma_h);
431  for (int i = 1; i < 3; ++i) {
432  int bytewidth = av_image_get_linesize(in->format, in->width, i);
433  av_image_copy_plane(out->data[i], out->linesize[i],
434  in->data[i], in->linesize[i],
435  bytewidth, uv_height);
436  }
437  } else {
438  sws_scale(ctx->sws_uv_scale, (const uint8_t **)(in->data + 1), in->linesize + 1,
439  0, ctx->sws_uv_height, out->data + 1, out->linesize + 1);
440  sws_scale(ctx->sws_uv_scale, (const uint8_t **)(in->data + 2), in->linesize + 2,
441  0, ctx->sws_uv_height, out->data + 2, out->linesize + 2);
442  }
443 
444  return 0;
445 }
446 
448 {
449  AVFilterContext *context = inlink->dst;
450  AVFilterLink *outlink = context->outputs[0];
451  DnnProcessingContext *ctx = context->priv;
452  DNNReturnType dnn_result;
453  AVFrame *out;
454 
455  copy_from_frame_to_dnn(ctx, in);
456 
457  dnn_result = (ctx->dnn_module->execute_model)(ctx->model, &ctx->output, 1);
458  if (dnn_result != DNN_SUCCESS){
459  av_log(ctx, AV_LOG_ERROR, "failed to execute model\n");
460  av_frame_free(&in);
461  return AVERROR(EIO);
462  }
463 
464  out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
465  if (!out) {
466  av_frame_free(&in);
467  return AVERROR(ENOMEM);
468  }
469 
470  av_frame_copy_props(out, in);
471  copy_from_dnn_to_frame(ctx, out);
472 
473  if (isPlanarYUV(in->format))
474  copy_uv_planes(ctx, out, in);
475 
476  av_frame_free(&in);
477  return ff_filter_frame(outlink, out);
478 }
479 
481 {
483 
486  sws_freeContext(context->sws_uv_scale);
487 
488  if (context->dnn_module)
489  (context->dnn_module->free_model)(&context->model);
490 
491  av_freep(&context->dnn_module);
492 }
493 
495  {
496  .name = "default",
497  .type = AVMEDIA_TYPE_VIDEO,
498  .config_props = config_input,
499  .filter_frame = filter_frame,
500  },
501  { NULL }
502 };
503 
505  {
506  .name = "default",
507  .type = AVMEDIA_TYPE_VIDEO,
508  .config_props = config_output,
509  },
510  { NULL }
511 };
512 
514  .name = "dnn_processing",
515  .description = NULL_IF_CONFIG_SMALL("Apply DNN processing filter to the input."),
516  .priv_size = sizeof(DnnProcessingContext),
517  .init = init,
518  .uninit = uninit,
520  .inputs = dnn_processing_inputs,
521  .outputs = dnn_processing_outputs,
522  .priv_class = &dnn_processing_class,
523 };
void * model
Definition: dnn_interface.h:45
#define NULL
Definition: coverity.c:32
int av_image_get_linesize(enum AVPixelFormat pix_fmt, int width, int plane)
Compute the size of an image line with format pix_fmt and width width for the plane plane...
Definition: imgutils.c:76
static enum AVPixelFormat pix_fmt
Buffered I/O operations.
const AVPixFmtDescriptor * av_pix_fmt_desc_get(enum AVPixelFormat pix_fmt)
Definition: pixdesc.c:2573
This structure describes decoded (raw) audio or video data.
Definition: frame.h:308
#define SWS_BICUBIC
Definition: swscale.h:60
AVOption.
Definition: opt.h:248
static const AVOption dnn_processing_options[]
void(* free_model)(DNNModel **model)
Definition: dnn_interface.h:61
planar YUV 4:4:4, 24bpp, (1 Cr & Cb sample per 1x1 Y samples)
Definition: pixfmt.h:71
misc image utilities
Main libavfilter public API header.
packed RGB 8:8:8, 24bpp, RGBRGB...
Definition: pixfmt.h:68
const char * desc
Definition: nvenc.c:87
struct SwsContext * sws_uv_scale
#define OFFSET(x)
int channels
Definition: dnn_interface.h:40
static const AVFilterPad dnn_processing_outputs[]
struct SwsContext * sws_grayf32_to_gray8
AVFrame * ff_get_video_buffer(AVFilterLink *link, int w, int h)
Request a picture buffer with a specific set of permissions.
Definition: video.c:99
static av_cold void uninit(AVFilterContext *ctx)
uint8_t log2_chroma_w
Amount to shift the luma width right to find the chroma width.
Definition: pixdesc.h:92
AVFilterFormats * ff_make_format_list(const int *fmts)
Create a list of supported formats.
Definition: formats.c:283
const char * name
Pad name.
Definition: internal.h:60
AVFilterLink ** inputs
array of pointers to input links
Definition: avfilter.h:346
#define av_assert0(cond)
assert() equivalent, that is always enabled.
Definition: avassert.h:37
DNNModel *(* load_model)(const char *model_filename)
Definition: dnn_interface.h:57
AVFilter ff_vf_dnn_processing
int ff_filter_frame(AVFilterLink *link, AVFrame *frame)
Send a frame of data to the next filter.
Definition: avfilter.c:1091
uint8_t
#define av_cold
Definition: attributes.h:88
AVOptions.
static int check_modelinput_inlink(const DNNData *model_input, const AVFilterLink *inlink)
struct SwsContext * sws_getContext(int srcW, int srcH, enum AVPixelFormat srcFormat, int dstW, int dstH, enum AVPixelFormat dstFormat, int flags, SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
Allocate and return an SwsContext.
Definition: utils.c:1911
static int config_input(AVFilterLink *inlink)
DNNBackendType
Definition: dnn_interface.h:33
external API header
#define av_log(a,...)
DNNDataType
Definition: dnn_interface.h:35
A filter pad used for either input or output.
Definition: internal.h:54
static int copy_from_frame_to_dnn(DnnProcessingContext *ctx, const AVFrame *frame)
DNNReturnType(* get_input)(void *model, DNNData *input, const char *input_name)
Definition: dnn_interface.h:48
int width
Definition: frame.h:366
#define AV_LOG_ERROR
Something went wrong and cannot losslessly be recovered.
Definition: log.h:194
int ff_set_common_formats(AVFilterContext *ctx, AVFilterFormats *formats)
A helper for query_formats() which sets all links to the same list of formats.
Definition: formats.c:600
uint8_t log2_chroma_h
Amount to shift the luma height right to find the chroma height.
Definition: pixdesc.h:101
int height
Definition: dnn_interface.h:40
static av_cold int init(AVFilterContext *context)
#define AV_PIX_FMT_FLAG_RGB
The pixel format contains RGB-like data (as opposed to YUV/grayscale).
Definition: pixdesc.h:148
void av_frame_free(AVFrame **frame)
Free the frame and any dynamically allocated objects in it, e.g.
Definition: frame.c:203
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification. ...
Definition: internal.h:153
void * data
Definition: dnn_interface.h:38
void * priv
private data for use by the filter
Definition: avfilter.h:353
simple assert() macros that are a bit more flexible than ISO C assert().
planar YUV 4:2:2, 16bpp, (1 Cr & Cb sample per 2x1 Y samples)
Definition: pixfmt.h:70
uint64_t flags
Combination of AV_PIX_FMT_FLAG_...
Definition: pixdesc.h:106
uint8_t nb_components
The number of components each pixel has, (1-4)
Definition: pixdesc.h:83
static int copy_uv_planes(DnnProcessingContext *ctx, AVFrame *out, const AVFrame *in)
AVFormatContext * ctx
Definition: movenc.c:48
void sws_freeContext(struct SwsContext *swsContext)
Free the swscaler context swsContext.
Definition: utils.c:2331
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame For filters that do not use the this method is called when a frame is wanted on an output For a it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return or at least make progress towards producing a frame
packed RGB 8:8:8, 24bpp, BGRBGR...
Definition: pixfmt.h:69
static const AVFilterPad dnn_processing_inputs[]
static int prepare_sws_context(AVFilterLink *outlink)
AVFILTER_DEFINE_CLASS(dnn_processing)
DNN inference engine interface.
DNNReturnType
Definition: dnn_interface.h:31
static const AVFilterPad outputs[]
Definition: af_acontrast.c:203
int format
format of the frame, -1 if unknown or unset Values correspond to enum AVPixelFormat for video frames...
Definition: frame.h:381
#define AV_PIX_FMT_GRAYF32
Definition: pixfmt.h:431
int linesize[AV_NUM_DATA_POINTERS]
For video, size in bytes of each picture line.
Definition: frame.h:339
Descriptor that unambiguously describes how the bits of a pixel are stored in the up to 4 data planes...
Definition: pixdesc.h:81
int attribute_align_arg sws_scale(struct SwsContext *c, const uint8_t *const srcSlice[], const int srcStride[], int srcSliceY, int srcSliceH, uint8_t *const dst[], const int dstStride[])
swscale wrapper, so we don&#39;t need to export the SwsContext.
Definition: swscale.c:744
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several inputs
uint8_t pi<< 24) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi-0x80)*(1.0f/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_U8, uint8_t,(*(const uint8_t *) pi-0x80)*(1.0/(1<< 7))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S16, int16_t,(*(const int16_t *) pi >> 8)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S16, int16_t,*(const int16_t *) pi *(1.0f/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S16, int16_t,*(const int16_t *) pi *(1.0/(1<< 15))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_S32, int32_t,(*(const int32_t *) pi >> 24)+0x80) CONV_FUNC_GROUP(AV_SAMPLE_FMT_FLT, float, AV_SAMPLE_FMT_S32, int32_t,*(const int32_t *) pi *(1.0f/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_DBL, double, AV_SAMPLE_FMT_S32, int32_t,*(const int32_t *) pi *(1.0/(1U<< 31))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_FLT, float, av_clip_uint8(lrintf(*(const float *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, float, av_clip_int16(lrintf(*(const float *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, float, av_clipl_int32(llrintf(*(const float *) pi *(1U<< 31)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_U8, uint8_t, AV_SAMPLE_FMT_DBL, double, av_clip_uint8(lrint(*(const double *) pi *(1<< 7))+0x80)) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, double, av_clip_int16(lrint(*(const double *) pi *(1<< 15)))) CONV_FUNC_GROUP(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, double, av_clipl_int32(llrint(*(const double *) pi *(1U<< 31))))#define SET_CONV_FUNC_GROUP(ofmt, ifmt) static void set_generic_function(AudioConvert *ac){}void ff_audio_convert_free(AudioConvert **ac){if(!*ac) return;ff_dither_free(&(*ac) ->dc);av_freep(ac);}AudioConvert *ff_audio_convert_alloc(AVAudioResampleContext *avr, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt, int channels, int sample_rate, int apply_map){AudioConvert *ac;int in_planar, out_planar;ac=av_mallocz(sizeof(*ac));if(!ac) return NULL;ac->avr=avr;ac->out_fmt=out_fmt;ac->in_fmt=in_fmt;ac->channels=channels;ac->apply_map=apply_map;if(avr->dither_method!=AV_RESAMPLE_DITHER_NONE &&av_get_packed_sample_fmt(out_fmt)==AV_SAMPLE_FMT_S16 &&av_get_bytes_per_sample(in_fmt) > 2){ac->dc=ff_dither_alloc(avr, out_fmt, in_fmt, channels, sample_rate, apply_map);if(!ac->dc){av_free(ac);return NULL;}return ac;}in_planar=ff_sample_fmt_is_planar(in_fmt, channels);out_planar=ff_sample_fmt_is_planar(out_fmt, channels);if(in_planar==out_planar){ac->func_type=CONV_FUNC_TYPE_FLAT;ac->planes=in_planar?ac->channels:1;}else if(in_planar) ac->func_type=CONV_FUNC_TYPE_INTERLEAVE;else ac->func_type=CONV_FUNC_TYPE_DEINTERLEAVE;set_generic_function(ac);if(ARCH_AARCH64) ff_audio_convert_init_aarch64(ac);if(ARCH_ARM) ff_audio_convert_init_arm(ac);if(ARCH_X86) ff_audio_convert_init_x86(ac);return ac;}int ff_audio_convert(AudioConvert *ac, AudioData *out, AudioData *in){int use_generic=1;int len=in->nb_samples;int p;if(ac->dc){av_log(ac->avr, AV_LOG_TRACE,"%d samples - audio_convert: %s to %s (dithered)\n", len, av_get_sample_fmt_name(ac->in_fmt), av_get_sample_fmt_name(ac->out_fmt));return ff_convert_dither(ac-> in
planar YUV 4:1:0, 9bpp, (1 Cr & Cb sample per 4x4 Y samples)
Definition: pixfmt.h:72
Describe the class of an AVClass context structure.
Definition: log.h:67
Filter definition.
Definition: avfilter.h:144
static int query_formats(AVFilterContext *context)
static int copy_from_dnn_to_frame(DnnProcessingContext *ctx, AVFrame *frame)
static av_always_inline int isPlanarYUV(enum AVPixelFormat pix_fmt)
const char * name
Filter name.
Definition: avfilter.h:148
AVFilterLink ** outputs
array of pointers to output links
Definition: avfilter.h:350
static enum AVPixelFormat pix_fmts[]
Definition: libkvazaar.c:275
DNNModule * ff_get_dnn_module(DNNBackendType backend_type)
Definition: dnn_interface.c:32
DNNReturnType(* execute_model)(const DNNModel *model, DNNData *outputs, uint32_t nb_output)
Definition: dnn_interface.h:59
uint8_t * data[AV_NUM_DATA_POINTERS]
pointer to the picture/channel planes.
Definition: frame.h:322
The exact code depends on how similar the blocks are and how related they are to the and needs to apply these operations to the correct inlink or outlink if there are several Macros are available to factor that when no extra processing is inlink
planar YUV 4:2:0, 12bpp, (1 Cr & Cb sample per 2x2 Y samples)
Definition: pixfmt.h:66
Y , 8bpp.
Definition: pixfmt.h:74
static int filter_frame(AVFilterLink *inlink, AVFrame *in)
planar YUV 4:1:1, 12bpp, (1 Cr & Cb sample per 4x1 Y samples)
Definition: pixfmt.h:73
#define LOG_FORMAT_CHANNEL_MISMATCH()
A list of supported formats for one end of a filter link.
Definition: formats.h:64
An instance of a filter.
Definition: avfilter.h:338
and forward the result(frame or status change) to the corresponding input.If nothing is possible
int height
Definition: frame.h:366
FILE * out
Definition: movenc.c:54
struct SwsContext * sws_gray8_to_grayf32
#define av_freep(p)
DNNReturnType(* set_input_output)(void *model, DNNData *input, const char *input_name, const char **output_names, uint32_t nb_output)
Definition: dnn_interface.h:51
#define av_always_inline
Definition: attributes.h:45
DNNBackendType backend_type
void av_image_copy_plane(uint8_t *dst, int dst_linesize, const uint8_t *src, int src_linesize, int bytewidth, int height)
Copy image plane from src to dst.
Definition: imgutils.c:338
const char * av_get_pix_fmt_name(enum AVPixelFormat pix_fmt)
Return the short name for a pixel format, NULL in case pix_fmt is unknown.
Definition: pixdesc.c:2489
internal API functions
static int config_output(AVFilterLink *outlink)
Filter the word “frame” indicates either a video frame or a group of audio as stored in an AVFrame structure Format for each input and each output the list of supported formats For video that means pixel format For audio that means channel sample they are references to shared objects When the negotiation mechanism computes the intersection of the formats supported at each end of a all references to both lists are replaced with a reference to the intersection And when a single format is eventually chosen for a link amongst the remaining all references to the list are updated That means that if a filter requires that its input and output have the same format amongst a supported all it has to do is use a reference to the same list of formats query_formats can leave some formats unset and return AVERROR(EAGAIN) to cause the negotiation mechanism toagain later.That can be used by filters with complex requirements to use the format negotiated on one link to set the formats supported on another.Frame references ownership and permissions
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default minimum maximum flags name is the option keep it simple and lowercase description are in without and describe what they for example set the foo of the bar offset is the offset of the field in your local context
DNNDataType dt
Definition: dnn_interface.h:39
AVPixelFormat
Pixel format.
Definition: pixfmt.h:64
int av_frame_copy_props(AVFrame *dst, const AVFrame *src)
Copy only "metadata" fields from src to dst.
Definition: frame.c:659
int i
Definition: input.c:406
#define AV_CEIL_RSHIFT(a, b)
Definition: common.h:58
#define FLAGS
#define check(x, y, S, v)